vigthoria-cli 1.13.13 → 1.13.20

package/dist/utils/clientManifest.js

@@ -56,8 +56,18 @@ export function resolveClientOsSlug() {
 }
 export function buildClientManifest(overrides = {}) {
     const ios = isIOSEnvironment() && !overrides.os;
+    const productId = overrides.product_id || process.env.VIGTHORIA_SOURCE_PRODUCT_ID || 'vigthoria-cli';
     return {
-        product_id: overrides.product_id || 'vigthoria-cli',
+        product_id: productId,
+        context_contract_version: overrides.context_contract_version || '2.0',
+        source_product_id: overrides.source_product_id || process.env.VIGTHORIA_SOURCE_PRODUCT_ID || productId,
+        active_application_id: overrides.active_application_id || process.env.VIGTHORIA_ACTIVE_APPLICATION_ID || 'cli-agent-chat',
+        active_surface_id: overrides.active_surface_id || process.env.VIGTHORIA_ACTIVE_SURFACE_ID || 'agent-session',
+        target_product_id: overrides.target_product_id || 'none',
+        workspace_kind: overrides.workspace_kind || process.env.VIGTHORIA_WORKSPACE_KIND || 'local-development-workspace',
+        selected_engine_id: overrides.selected_engine_id || 'none',
+        capability_catalog_version: overrides.capability_catalog_version || process.env.VIGTHORIA_CAPABILITY_CATALOG_VERSION || '2026.08.03.1',
+        capability_scope_id: overrides.capability_scope_id || process.env.VIGTHORIA_CAPABILITY_SCOPE_ID || 'cli.ecosystem-user',
         version: overrides.version || readPackageVersion(),
         os: overrides.os || resolveClientOsSlug(),
         execution_mode: overrides.execution_mode || (ios ? 'mobile-terminal-sandbox' : 'local-user-machine'),

package/dist/utils/command-menu.js

@@ -3,6 +3,10 @@ const AGENT_COMMANDS = new Set([
     'agent', 'chat', 'chat-resume', 'operator', 'legion', 'workflow',
     'history', 'replay', 'fork', 'cancel',
 ]);
+// Subprocess protocols are not user-facing commands. Presenting one in the
+// root picker caused `v4-menu` to be launched without its required output
+// file, which is exactly the failure reported from CLI 1.13.13.
+const INTERNAL_COMMANDS = new Set(['v4-menu']);
 function familyFor(commandPath) {
     const root = commandPath.split(' ')[0];
     if (root === 'repo')
@@ -15,6 +19,8 @@ export function buildCommandCatalog(program) {
     const entries = [];
     const visit = (command, parents) => {
         const commandPath = [...parents, command.name()].join(' ');
+        if (parents.length === 0 && INTERNAL_COMMANDS.has(commandPath))
+            return;
         entries.push({
             path: commandPath,
             description: command.description() || '',

package/dist/utils/config.js

@@ -195,11 +195,13 @@ export class Config {
         // Vigthoria server infrastructure operational models
         // ═══════════════════════════════════════════════════════════════
         const models = [
-            { id: 'code', name: 'Vigthoria v3 Code 35B', description: 'Native 35B coding model on Blackwell (V3 pipeline)', tier: 'local', backendModel: 'vigthoria-v3-code-35b' },
-            { id: 'code-35b', name: 'Vigthoria v3 Code 35B', description: 'Same flagship model as code', tier: 'local', backendModel: 'vigthoria-v3-code-35b' },
-            { id: 'code-9b', name: 'Vigthoria v3 Code 9B', description: 'Efficient coding specialist for quick tasks', tier: 'local', backendModel: 'vigthoria-v3-code-9b' },
-            { id: 'balanced', name: 'Vigthoria Balanced 4B', description: 'Balanced general-purpose local model', tier: 'local', backendModel: 'vigthoria-v3-balanced-4b' },
-            { id: 'balanced-4b', name: 'Vigthoria Balanced 4B', description: 'Efficient 4B general-purpose model (Qwen3.5-4B based)', tier: 'local', backendModel: 'vigthoria-v3-balanced-4b' },
+            { id: 'architect', name: 'Vigthoria v4 Creative 27B', description: 'Architect, planning, deep reasoning, and creative systems work', tier: 'local', backendModel: 'Vigthoria-v4-Creative-27B' },
+            { id: 'code', name: 'Vigthoria v4 Code 27B', description: 'Dense production executor for coding, debugging, and repository changes', tier: 'local', backendModel: 'Vigthoria-v4-Code-27B' },
+            { id: 'assistant', name: 'Vigthoria v4 Assistant 9B', description: 'Low-latency FIM, completion, voice, and diagnostic model', tier: 'local', backendModel: 'Vigthoria-v4-Assistant-9B' },
+            { id: 'code-35b', name: 'Vigthoria v4 Code 27B', description: 'Legacy selector mapped to the V4 executor', tier: 'local', backendModel: 'Vigthoria-v4-Code-27B' },
+            { id: 'code-9b', name: 'Vigthoria v4 Assistant 9B', description: 'Legacy selector mapped to the V4 assistant', tier: 'local', backendModel: 'Vigthoria-v4-Assistant-9B' },
+            { id: 'balanced', name: 'Vigthoria v4 Assistant 9B', description: 'Legacy selector mapped to the V4 assistant', tier: 'local', backendModel: 'Vigthoria-v4-Assistant-9B' },
+            { id: 'balanced-4b', name: 'Vigthoria v4 Assistant 9B', description: 'Legacy selector mapped to the V4 assistant', tier: 'local', backendModel: 'Vigthoria-v4-Assistant-9B' },
         ];
         if (this.hasCloudAccess()) {
             models.push({ id: 'cloud-fast', name: 'Vigthoria Cloud Fast', description: 'Fast cloud responses for lighter work', tier: 'cloud', backendModel: 'vigthoria-cloud-fast' }, { id: 'cloud-balanced', name: 'Vigthoria Cloud Balanced', description: 'Default quality/cost balance for chat and coding', tier: 'cloud', backendModel: 'vigthoria-cloud-balanced' }, { id: 'cloud-code', name: 'Vigthoria Cloud Code', description: 'Economical cloud coding and completion', tier: 'cloud', backendModel: 'vigthoria-cloud-code' }, { id: 'cloud-power', name: 'Vigthoria Cloud Power', description: 'Premium general intelligence for demanding work', tier: 'cloud', backendModel: 'vigthoria-cloud-power' }, { id: 'cloud-maximum', name: 'Vigthoria Cloud Maximum', description: 'Maximum power for complex architecture and reviews', tier: 'cloud', backendModel: 'vigthoria-cloud-maximum' }, 

package/dist/utils/contextBudget.d.ts

@@ -15,7 +15,11 @@ export declare function estimateJsonTokens(value: unknown): number;
 export declare function resolveInputBudget(runtimeNctx?: number, maxOutput?: number): number;
 /**
  * Char limit for the initial V3 context JSON payload.
- * ~45% of input token budget × 4 chars/token, clamped 32k–100k.
+ * The hydration payload is transport-only: the server writes workspaceFiles
+ * to an isolated directory and exposes only a compact filename summary to the
+ * model. It can therefore use the full calculated input envelope without
+ * consuming system/tool framing tokens. Clamp at 180k, far below the gateway
+ * body limit, while preserving compaction for genuinely large workspaces.
  */
 export declare function resolveV3ContextCharLimit(runtimeNctx?: number): number;
 export declare function compactToolListingOutput(name: string, output: string, charCap: number): string;

package/dist/utils/contextBudget.js

@@ -29,13 +29,16 @@ export function resolveInputBudget(runtimeNctx = DEFAULT_CODE_RUNTIME_CTX, maxOu
 }
 /**
  * Char limit for the initial V3 context JSON payload.
- * ~45% of input token budget × 4 chars/token, clamped 32k–100k.
+ * The hydration payload is transport-only: the server writes workspaceFiles
+ * to an isolated directory and exposes only a compact filename summary to the
+ * model. It can therefore use the full calculated input envelope without
+ * consuming system/tool framing tokens. Clamp at 180k, far below the gateway
+ * body limit, while preserving compaction for genuinely large workspaces.
  */
 export function resolveV3ContextCharLimit(runtimeNctx = DEFAULT_CODE_RUNTIME_CTX) {
     const inputBudget = resolveInputBudget(runtimeNctx);
-    const initialTurnTokens = Math.floor(inputBudget * 0.45);
-    const charLimit = initialTurnTokens * CHARS_PER_TOKEN_EST;
-    return Math.min(100_000, Math.max(32_000, charLimit));
+    const hydrationChars = inputBudget * CHARS_PER_TOKEN_EST;
+    return Math.min(180_000, Math.max(32_000, hydrationChars));
 }
 export function compactToolListingOutput(name, output, charCap) {
     const text = String(output || '');

package/dist/utils/fastAgentRouter.d.ts

@@ -1,5 +1,5 @@
 /**
- * Vigthoria Balanced 4B — fast agent path classifier (~0.7–2s on local inference :8016).
+ * Vigthoria v4 Assistant 9B — fast agent path classifier through the governed router.
  * Regex/trivial detection always wins; LLM router refines ambiguous build prompts only.
  */
 export type AgentRoutePath = 'template-instant' | 'v3-agent' | 'analysis-only';
@@ -23,7 +23,7 @@ export interface BalancedRouterResult {
 }
 export declare function parseBalancedRouterJson(raw: string): BalancedRouterJson | null;
 /**
- * Inference bases for Balanced 4B routing only.
+ * Inference bases for the V4 Assistant routing lane only.
  * Local developer machines must opt in (env) — avoids ~1.5s localhost probe on every Windows turn.
  */
 export declare function getInferenceRouterBaseUrls(selfHostedUrl: string | null): string[];

package/dist/utils/fastAgentRouter.js

@@ -1,5 +1,5 @@
 /**
- * Vigthoria Balanced 4B — fast agent path classifier (~0.7–2s on local inference :8016).
+ * Vigthoria v4 Assistant 9B — fast agent path classifier through the governed router.
  * Regex/trivial detection always wins; LLM router refines ambiguous build prompts only.
  */
 const ROUTER_SYSTEM = `You are Vigthoria Agent Router. Reply with ONE JSON object only. No markdown, no HTML, no code fences.
@@ -14,8 +14,7 @@ fast_path: "template-instant" or null
 
 Example: {"path":"template-instant","task_kind":"web-build","fast_path":"template-instant","confidence":0.95,"reason":"trivial html popup page"}`;
 const INFERENCE_MODEL_IDS = [
-    'vigthoria-balanced-4b',
-    'vigthoria-v3-balanced-4b',
+    'Vigthoria-v4-Assistant-9B',
 ];
 function normalizePath(value) {
     const p = String(value || '').trim().toLowerCase();
@@ -56,7 +55,7 @@ export function parseBalancedRouterJson(raw) {
     return null;
 }
 /**
- * Inference bases for Balanced 4B routing only.
+ * Inference bases for the V4 Assistant routing lane only.
  * Local developer machines must opt in (env) — avoids ~1.5s localhost probe on every Windows turn.
  */
 export function getInferenceRouterBaseUrls(selfHostedUrl) {
@@ -69,7 +68,7 @@ export function getInferenceRouterBaseUrls(selfHostedUrl) {
         || process.env.VIGTHORIA_AGENT_LLM_ROUTER === '1'
         || process.platform === 'linux';
     if (allowLocalhost) {
-        urls.push('http://127.0.0.1:8016');
+        urls.push('http://127.0.0.1:4009');
     }
     return [...new Set(urls)];
 }

package/dist/utils/requestIntent.js

@@ -4,14 +4,15 @@
  */
 const CLI_SHAPING_BLOCK = /(?:^|\n\n)Platform:\s*(?:Windows|macOS|Linux)\.[\s\S]*?(?=\n\n[A-Z][^\n]{0,40}:|\s*$)/i;
 const READONLY_SHAPING_BLOCK = /(?:^|\n\n)(?:Read-only analysis mode is active\.|Diagnostic mode is active\.)[\s\S]*?(?=\n\n[A-Z][^\n]{0,40}:|\s*$)/i;
-const BUILD_VERBS = /\b(build|create|make|implement|complete|fix|repair|edit|modify|write|generate|add|finish|scaffold|develop|update|change|refactor|want|need|require|looking for|let's|let us|we need|we should|erstelle|erstellen|schreib|schreibe|bearbeite)\b/i;
-const ARTIFACT_NOUNS = /\b(file|files|project|game|spiel|app|website|html5|frontend|component|feature|code|workspace|repo|clone|replica|remake|page|site|screen|button|popup|modal|alert|canvas|sprite|level|levels|enemy|enemies|character|npc|world|scene|script|module|api|database|config|landing|dashboard|platformer|arcade|playable|collectible|scoreboard|leaderboard|pitfall|pacman|rogue|html|css|javascript|typescript)\b/i;
+const BUILD_VERBS = /\b(build|create|make|implement|complete|redesign|overhaul|rebuild|transform|polish|art-direct|fix|repair|edit|modify|write|generate|add|finish|scaffold|develop|update|change|refactor|want|need|require|looking for|let's|let us|we need|we should|erstelle|erstellen|schreib|schreibe|bearbeite)\b/i;
+const ARTIFACT_NOUNS = /\b(file|files|project|game|spiel|app|website|html5|frontend|component|feature|code|workspace|repo|clone|replica|remake|page|site|design|showcase|screen|button|popup|modal|alert|canvas|sprite|level|levels|enemy|enemies|character|npc|world|scene|script|module|api|database|config|landing|dashboard|platformer|arcade|playable|collectible|scoreboard|leaderboard|pitfall|pacman|rogue|html|css|javascript|typescript)\b/i;
 /** Canvas/playable games only — bare "html5" or "website" must NOT match. */
 const GAME_INTENT = /\b(game|spiel|playable|babylon(?:\.js|js)?|vge|webgl|3d\s+(?:game|runner|prototype|world|experience)|html5\s+(?:game|canvas)|html5\s+game|canvas\s+game|game\s+canvas|arcade|platformer|side[- ]?scroller|pitfall|pac[- ]?man|tetris|snake|breakout|pong|roguelike|metroidvania|tower\s+defense|clone\s+of\s+(?:a\s+)?(?:game|pitfall|pac|mario|zelda|arcade)|gameplay|collectible|boss\s+fight|level\s+design|sprite\s+sheet|wild\s+(?:boar|pig|forest)|forest\s+pig)\b/i;
 const WEB_PAGE_INTENT = /\b(website|web\s*site|webpage|web\s*page|landing\s+page|home\s*page|index\.html|html\s+page|static\s+page|single\s+page|popup|alert|modal|hello\s+world)\b/i;
+const WEB_SHOWCASE_INTENT = /\b(web[- ]?design\s+showcase|frontend|marketing\s+site|dashboard|saas)\b/i;
 const CLONE_BUILD = /\bclone\s+of\b|\b(build|create|make)\s+(?:a|an|the|us|me)?\s*(?:clone|replica|remake|port)\b/i;
 const READ_ONLY_INTENT = /\b(analy[sz]e|analyse|analysis|audit|review|inspect|proof|understand|summari[sz]e|scan|read[\s-]?only|where we left|what is|how does|show me|tell me|identify\s+gaps?|gap\s+analysis|production\s+blockers?)\b/i;
-const EXPLICIT_WRITE_VERBS = /\b(implement|write|edit|modify|update|fix|repair|create|build|generate|scaffold|deploy|refactor|develop|integrate|correct|apply|patch)\b/i;
+const EXPLICIT_WRITE_VERBS = /\b(implement|write|edit|modify|update|redesign|overhaul|rebuild|transform|polish|art-direct|fix|repair|create|build|generate|scaffold|deploy|refactor|develop|integrate|correct|apply|patch)\b/i;
 const TRIVIAL_HTML_PAGE = /\b(hello\s*world|simple\s+(?:html|web|page|site)|html5\s+website|html\s+5\s+website|popup|alert\s*\(|show\s+a\s+popup|one\s+page|single\s+html)\b/i;
 /** User explicitly grants write permission after an analysis or confirmation gate. */
 const WRITE_PERMISSION_GRANT = /\b(?:hereby\s+confirm|i\s+confirm|confirm(?:ed)?\s+that|give\s+(?:you|permission)|granting\s+permission|allow\s+(?:you|me\s+to)|you\s+(?:may|can|have\s+permission\s+to)|permission\s+to|authorized?\s+to|approve(?:d)?\s+(?:the\s+)?(?:changes?|writes?|edits?)|go\s+ahead\s+and(?:\s+write|\s+apply|\s+make|\s+do)?|do\s+all\s+the\s+changes|apply\s+(?:the\s+)?(?:changes?|fixes?|fix)|proceed\s+with\s+(?:the\s+)?(?:changes?|fixes?|implementation|writes?)|make\s+(?:the\s+)?changes|please\s+write)\b/i;
@@ -111,7 +112,7 @@ export function isContinueOrRetryPrompt(prompt) {
 }
 export function hasWebPageIntent(prompt) {
     const text = stripExecutionShaping(prompt);
-    return WEB_PAGE_INTENT.test(text) || /\bhtml5\s+website\b/i.test(text);
+    return WEB_PAGE_INTENT.test(text) || WEB_SHOWCASE_INTENT.test(text) || /\bhtml5\s+website\b/i.test(text);
 }
 export function isTrivialHtmlPageRequest(prompt) {
     const text = stripExecutionShaping(prompt);

package/dist/utils/session.d.ts

@@ -47,6 +47,24 @@ export interface Session {
     agentMode: boolean;
     operatorMode?: boolean;
     lastAgentRunOutcome?: Record<string, unknown> | null;
+    activeAgentRun?: AgentExecutionCheckpoint | null;
+}
+export interface AgentExecutionCheckpoint {
+    executionId: string;
+    parentExecutionId?: string | null;
+    contextId: string;
+    prompt: string;
+    originalPrompt: string;
+    workspacePath: string;
+    userTier: string | null;
+    taskType: string;
+    workflowType: string;
+    status: 'submitted' | 'running' | 'completed' | 'failed';
+    submittedAt: string;
+    updatedAt: string;
+    failedTaskIds: string[];
+    unfinishedTaskIds: string[];
+    error?: string | null;
 }
 export declare class SessionManager {
     private sessionsDir;

package/dist/utils/v3-workspace-path.js

@@ -4,7 +4,7 @@ function stripInlineWorkspacePlaceholders(value) {
     // Only strip placeholders when they were appended to a segment (e.g. pitfall[Vigthoria service]).
     // Keep canonical scrubbed segments like /[internal]/ intact for dedicated mapping logic.
     return String(value || '')
-        .replace(/(^|[^/])\[(?:internal|Vigthoria service|Vigthoria Agent|Vigthoria storage|temp|home|workspace)\](?=(?:\/|$))/ig, '$1')
+        .replace(/([^/])\[(?:internal|Vigthoria service|Vigthoria Agent|Vigthoria storage|temp|home|workspace)\](?=(?:\/|$))/ig, '$1')
         .replace(/\/+$/g, '');
 }
 /**
@@ -16,6 +16,9 @@ export function normalizeV3WorkspaceRelativePath(rawPath, rootPath) {
     if (!input) {
         return '';
     }
+    input = input
+        .replace(/^var\/?\[internal\]/i, '[internal]')
+        .replace(/^varwww\[internal\]/i, '[internal]');
     input = stripInlineWorkspacePlaceholders(input);
     const safeRelative = (candidate) => {
         const stripped = String(candidate || '').replace(/^\/+/, '');
@@ -64,6 +67,8 @@ export function normalizeV3WorkspaceRelativePath(rawPath, rootPath) {
         }
         const placeholderPatterns = [
             [/^(?:var\/)?www\/\[internal\](?:\/(.*))?$/i, 1],
+            [/^var\/?\[internal\](?:\/(.*))?$/i, 1],
+            [/^varwww\[internal\](?:\/(.*))?$/i, 1],
             [/^\[internal\](?:\/(.*))?$/i, 1],
             [/^\[Vigthoria service\](?:\/(.*))?$/i, 1],
             [/^\[Vigthoria Agent\](?:\/(.*))?$/i, 1],
@@ -134,6 +139,9 @@ export function normalizeV3WorkspaceRelativePath(rawPath, rootPath) {
         }
         const rootNoLeadingSlash = normalizedRoot.replace(/^\//, '');
         const rootBase = path.posix.basename(normalizedRoot);
+        if (input.replace(/\/+$/g, '').toLowerCase() === normalizedRoot.toLowerCase()) {
+            return '';
+        }
         const prefixes = [
             `${normalizedRoot}/`,
             `${rootNoLeadingSlash}/`,
@@ -150,6 +158,11 @@ export function normalizeV3WorkspaceRelativePath(rawPath, rootPath) {
             return safeRelative(input.slice(embeddedIndex + embeddedRoot.length));
         }
     }
+    // An unmatched POSIX absolute path belongs to another filesystem boundary.
+    // Never reinterpret it as a workspace-relative path by merely trimming '/'.
+    if (input.startsWith('/')) {
+        return '';
+    }
     const relative = safeRelative(input);
     if (relative === 'workspace') {
         return '';
@@ -167,7 +180,11 @@ export function isV3WorkspaceRootAliasPath(rawPath, rootPath) {
     if (!original || original === '.') {
         return true;
     }
-    const sanitized = stripInlineWorkspacePlaceholders(original).replace(/\\/g, '/').trim();
+    const compactCanonical = original
+        .replace(/\\/g, '/')
+        .replace(/^var\/?\[internal\]/i, '[internal]')
+        .replace(/^varwww\[internal\]/i, '[internal]');
+    const sanitized = stripInlineWorkspacePlaceholders(compactCanonical).trim();
     if (!sanitized || sanitized === '.' || sanitized === '/') {
         return true;
     }
@@ -181,6 +198,9 @@ export function isV3WorkspaceRootAliasPath(rawPath, rootPath) {
         || lowered === '[internal]'
         || lowered === 'www/[internal]'
         || lowered === 'var/www/[internal]'
+        || lowered === 'var[internal]'
+        || lowered === 'var/[internal]'
+        || lowered === 'varwww[internal]'
         || lowered === '[vigthoria service]'
         || lowered === 'var/www/[vigthoria service]') {
         return true;

package/install.ps1

@@ -5,7 +5,7 @@
 $ErrorActionPreference = "Stop"
 
 # Configuration
-$CLI_VERSION = "1.13.13"
+$CLI_VERSION = "1.13.20"
 $INSTALL_DIR = "$env:USERPROFILE\.vigthoria"
 $NPM_PACKAGE = "vigthoria-cli"
 $GIT_PACKAGE_URL = "git+https://market.vigthoria.io/vigthoria/vigthoria-cli.git"

package/install.sh

@@ -26,7 +26,7 @@ else
 fi
 
 # Configuration
-CLI_VERSION="1.13.13"
+CLI_VERSION="1.13.20"
 INSTALL_DIR="$HOME/.vigthoria"
 REPO_URL="https://market.vigthoria.io/vigthoria/vigthoria-cli"
 GIT_PACKAGE_URL="git+https://market.vigthoria.io/vigthoria/vigthoria-cli.git"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vigthoria-cli",
-  "version": "1.13.13",
+  "version": "1.13.20",
   "description": "Vigthoria Coder CLI - AI-powered terminal coding assistant",
   "type": "module",
   "main": "dist/index.js",
@@ -72,7 +72,9 @@
     "test:legion:billing:e2e": "npm run build && node scripts/test-legion-godmode-billing-e2e.js",
     "test:windows:v3-sync": "npm run build && node scripts/test-windows-v3-sync-recovery.js",
     "test:v3-workspace-path": "npm run build && node scripts/test-v3-workspace-path.js",
+    "test:v3-server-tool-execution": "npm run build && node scripts/test-v3-server-tool-execution.js",
     "test:session:project-match": "npm run build && node scripts/test-session-project-match.mjs",
+    "test:external-client-resume": "npm run build && node scripts/test-external-client-resume.mjs",
     "test:v3-stream-mutation": "npm run build && node scripts/test-v3-stream-mutation.js",
     "test:context:budget": "npm run build && node scripts/test-context-budget.js",
     "test:v3-client-tool-quality": "npm run build && node scripts/test-v3-client-tool-quality.mjs",

package/scripts/release/validate-no-go-gates.sh

@@ -5,6 +5,7 @@ ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
 cd "$ROOT"
 
 CLI="node dist/index.js"
+SERVICE_IDENTITY_MODE="${VIGTHORIA_TEST_SERVICE_IDENTITY:-0}"
 unset V3_SERVICE_KEY
 unset HYPERLOOP_SERVICE_KEY
 # Release validation runs against local model-router in this environment
@@ -27,6 +28,7 @@ printf '%s
 ' "$AUTH_JSON" > /tmp/vig-auth-preflight.json
 python3 - << 'PY'
 import json
+import re
 with open('/tmp/vig-auth-preflight.json','r',encoding='utf-8') as f:
     j=json.load(f)
 msg='Vigthoria Gate way user authentification failed. Please log out and login again.'
@@ -35,6 +37,15 @@ assert j.get('error') == msg, f"unexpected auth message: {j.get('error')}"
 print('[pass] auth preflight message')
 PY
 
+if [[ "$SERVICE_IDENTITY_MODE" == "1" ]]; then
+  # From this point onward the no-go suite exercises server-local model and
+  # tool routes without borrowing a human account session. The invalid-token
+  # gateway assertion above always runs against the canonical public boundary.
+  export VIGTHORIA_LOCAL_TEST_MODE=1
+  export VIGTHORIA_LOCAL_TEST_TOKEN="vigthoria-release-service-identity"
+  export VIGTHORIA_AUTH_TOKEN="$VIGTHORIA_LOCAL_TEST_TOKEN"
+fi
+
 echo "[1.1] ranker includes agent.py top10 for SSE prompt"
 node - << 'EOF2'
 import('./dist/utils/context-ranker.js').then((m)=>{
@@ -62,12 +73,22 @@ if (!tsc || !tsc.ran || !tsc.passed) process.exit(1);
 EOF2
 
 echo "[6.2] balanced-4b no-agent path"
-$CLI chat --no-agent --model balanced-4b --prompt "reply exactly: ok" --json >/tmp/vig-balanced.json
+for attempt in 1 2 3; do
+  if $CLI chat --no-agent --model balanced-4b --prompt "reply exactly: ok" --json >/tmp/vig-balanced.json; then
+    break
+  fi
+  if [[ "$attempt" == "3" ]]; then
+    echo "[fail] balanced-4b remained unavailable after 3 attempts"
+    exit 1
+  fi
+  echo "[retry] balanced-4b transient failure ($attempt/3)"
+  sleep 2
+done
 python3 - << 'PY'
 import json
 j=json.load(open('/tmp/vig-balanced.json'))
 assert j.get('success') is True
-assert j.get('model') in ('vigthoria-balanced-4b', 'vigthoria-v3-balanced-4b'), j
+assert j.get('model') in ('Vigthoria-v4-Assistant-9B', 'vigthoria-balanced-4b', 'vigthoria-v3-balanced-4b'), j
 print('[pass] balanced-4b')
 PY
 
@@ -76,13 +97,17 @@ $CLI hyper-loop status >/tmp/vig-hyperloop.txt
 $CLI devtools connect >/tmp/vig-devtools.txt
 
 echo "[6.11] operator flow"
-$CLI operator --prompt "reply with exactly: OK" --json >/tmp/vig-operator.json
-python3 - << 'PY'
+if [[ "$SERVICE_IDENTITY_MODE" == "1" ]]; then
+  echo "[skip] paid-user operator probe (isolated release service identity)"
+else
+  $CLI operator --prompt "reply with exactly: OK" --json >/tmp/vig-operator.json
+  python3 - << 'PY'
 import json
 j=json.load(open('/tmp/vig-operator.json'))
 assert j.get('success') is True, j
 print('[pass] operator flow')
 PY
+fi
 
 echo "[7.1/7.3/7.4] local service health"
 for u in http://localhost:8030/health http://localhost:4009/health http://localhost:4011/health; do
@@ -95,10 +120,12 @@ echo "[7.2/7.8/12.3] models inventory"
 curl -s http://localhost:4009/v1/models >/tmp/vig-models.json
 python3 - << 'PY'
 import json
+import re
 ids={m.get('id','') for m in json.load(open('/tmp/vig-models.json')).get('data',[])}
 if ids:
   has_router = any(
     m in ids for m in (
+      'Vigthoria-v4-Assistant-9B',
       'vigthoria-balanced-4b', 'vigthoria-balanced-4b:latest',
       'vigthoria-v3-balanced-4b', 'vigthoria-v3-balanced-4b:latest',
       'vigthoria-fast-9b', 'vigthoria-fast-9b:latest',
@@ -106,11 +133,12 @@ if ids:
   )
   has_creative = any(
     m in ids for m in (
+      'Vigthoria-v4-Creative-27B',
       'vigthoria-creative-9b-v4', 'vigthoria-creative-9b-v4:latest',
       'vigthoria-fast-9b', 'vigthoria-fast-9b:latest',
     )
   )
-  has_code = any('vigthoria-v3-code-35b' in i for i in ids)
+  has_code = 'Vigthoria-v4-Code-27B' in ids or any(re.search(r'^vigthoria-v3(?:\.\d+)?-code-35b(?:-|:|$)', i) for i in ids)
   assert has_router, f'missing router model in inventory: {sorted(ids)}'
   assert has_creative, f'missing creative/fast model in inventory: {sorted(ids)}'
   assert has_code, f'missing code model in inventory: {sorted(ids)}'
@@ -136,12 +164,22 @@ rg -n "coder.vigthoria.io/releases" install.ps1 install.sh README.md >/tmp/vig-r
 cat /tmp/vig-release-urls.txt
 
 echo "[12.1] no-agent governance fallback"
-$CLI chat --no-agent --model creative --prompt "say ok" --json >/tmp/vig-creative.json
+for attempt in 1 2 3; do
+  if $CLI chat --no-agent --model creative-v4 --prompt "say ok" --json >/tmp/vig-creative.json; then
+    break
+  fi
+  if [[ "$attempt" == "3" ]]; then
+    echo "[fail] governance fallback remained unavailable after 3 attempts"
+    exit 1
+  fi
+  echo "[retry] governance fallback transient failure ($attempt/3)"
+  sleep 2
+done
 python3 - << 'PY'
 import json
 j=json.load(open('/tmp/vig-creative.json'))
 assert j.get('success') is True
-assert j.get('model') == 'vigthoria-v3-code-35b', j
+assert j.get('model') == 'Vigthoria-v4-Code-27B', j
 md=j.get('metadata') or {}
 fb=md.get('modelFallback') or {}
 assert fb.get('reason') == 'governance-blocked-model', j